The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes external Python scripts using shell commands in Phase 2. Arguments for these scripts, such as --urls, --subreddit, --keywords, and --query, are interpolated directly from a JSON configuration file (clients/<client>/config/industry-scanner.json). If this configuration file is modified by a malicious actor, it could lead to arbitrary command execution on the host system.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests large volumes of untrusted data from various sources, including web searches, blogs, Reddit, Twitter/X, LinkedIn, Hacker News, RSS feeds, and review sites.
Boundary markers: The instructions do not define clear delimiters or "ignore instructions" warnings when processing the collected data during the consolidation and strategy generation phases.
Capability inventory: The skill has the ability to read and write files and execute shell commands (to run sub-scrapers).
Sanitization: There is no evidence of sanitization or filtering to remove hidden instructions or adversarial content from the scraped data before it is presented to the agent for analysis.
[DATA_EXFILTRATION]: The skill requires access to sensitive client files, specifically clients/<client>/context.md, which contains positioning, value propositions, and Ideal Customer Profiles (ICP). While intended for internal processing, the skill's capabilities (network access via scrapers and shell execution) represent a potential path for data exposure if the agent's logic is subverted by malicious input.

industry-scanner